What is incrementality in marketing? How to measure it, including with a data clean room

Incrementality in marketing: definition

Incrementality in marketing is the portion of an outcome, such as sales, sign-ups, or store visits, that a campaign caused and that would not have happened without it. The underlying question is what would have happened if the campaign had never run. That hypothetical outcome is called the counterfactual. It cannot be observed for the users a campaign reaches, but a control group shows it: a comparable set of users that never sees the ads.

Standard campaign reports cannot reveal what would have happened without the campaign. A report can show how many people saw an ad and later bought, but some of those people were already planning to buy. Loyal customers and people who search for a brand by name convert at high rates whether or not an ad reaches them. Counting each of those conversions as a result of the campaign overstates its impact.

Incrementality separates the conversions a campaign caused from the baseline that would have occurred anyway. A handful of terms recur in this kind of measurement:

• Baseline: the outcome that occurs without the marketing activity.
• Incremental lift: the difference between the outcome with the campaign and the baseline, usually expressed as a percentage.
• Incremental conversions or revenue: the absolute number of extra conversions, or the revenue they generated, attributable to the campaign.
• Incremental ROAS (iROAS): incremental revenue divided by ad spend.
• Cost per incremental acquisition (CPIA): ad spend divided by the number of incremental conversions.

Incrementality is measured with controlled experiments. One group is exposed to the marketing activity, a comparable control group is held out, and the difference in outcomes between the two is the incremental effect. The sections below cover how this differs from attribution, which test designs exist, and how the same logic works when the conversion data sits with a retailer or advertiser. In a Decentriq data clean room, a publisher's exposure data is matched with an advertiser's or retailer's conversion data, and only aggregated lift results are released.

The result informs budget decisions. Channels close to the point of purchase, such as branded search and retargeting, often report strong returns while adding little beyond the baseline. Channels with weak click signals, such as connected TV or audio, can show the opposite pattern. Only a test against a control group reveals which case applies.

Incrementality vs attribution

Attribution assigns credit for a conversion to the marketing touchpoints that preceded it, while incrementality measures whether the marketing changed the outcome at all.

Attribution models, including last click, multi-touch, and data-driven variants, work from observed customer paths. They report which channels and ads appeared before a conversion, which supports fast and granular campaign optimization. They cannot show whether the conversion would have happened without those touchpoints, because every customer in the data has a path and none has a counterfactual. A retargeting campaign can receive credit for thousands of conversions in an attribution report, while a holdout test shows that most of those buyers would have returned on their own.

Incrementality testing builds the counterfactual by design. A control group that does not see the campaign shows what happens without it, so the gap between the two groups reflects the campaign's causal effect. Tests require planning, time, and a control group, which means they run on selected channels and campaigns and do not provide continuous coverage.

Marketing mix modeling (MMM) is the other common approach. It uses aggregated historical data, such as spend and sales over time, to estimate each channel's contribution at a strategic level. Mature measurement programs often combine the approaches, using incrementality test results to check and calibrate attribution and MMM outputs.

How to measure incrementality: test designs

The four main test designs differ in how the test and control groups are formed, which determines the channels each one suits and the data it needs.

Holdout tests

A holdout test randomly splits a defined audience, such as a CRM list or a retargeting pool, into a test group that receives the campaign and a control group that does not. Random assignment means the two groups differ only in exposure, so the gap in outcomes reflects the campaign's effect. Holdouts work where the advertiser controls who is targeted, including email, CRM, paid social, and programmatic audiences. The cost is a small, temporary loss of exposure among the control group.

Geo tests

A geo test assigns regions, such as cities, states, or postal code areas, to test or control. The campaign runs in the test regions only, and sales trends in both sets of regions are compared over several weeks. Control regions are chosen to behave like the test regions before the campaign starts, either by matching markets or by building a weighted synthetic control from historical data. Geo tests suit channels without user-level identifiers, such as TV, radio, and out-of-home, and campaigns where privacy rules limit user tracking. Spillover between neighboring regions and a limited number of comparable regions are the main risks.

Conversion lift tests

Conversion lift tests are user-level experiments run by an advertising platform. The platform randomly assigns eligible users to an exposed group or a control group and reports the difference in conversion rates. Setup effort is low, and results arrive in a standard format. The scope is limited to the platform's own environment, and the methodology and data stay with the platform, so effects on other channels remain invisible.

Sales lift tests

A sales lift test measures the effect of advertising on purchases recorded in sales data, including in-store transactions. Exposed and unexposed groups are compared on purchase rate or revenue, using transaction records from a retailer, a loyalty program, or the advertiser's own point of sale. This design measures the outcome closest to revenue. It also depends on combining advertiser exposure data with retailer or publisher data, which is the situation the clean room section below addresses.

Design considerations

A reliable test settles several points before launch:

• A specific hypothesis and one primary KPI, such as incremental purchases.
• A sample size and duration large enough to detect the expected effect with statistical significance. Small audiences or short tests can miss real lift.
• Clean separation between groups. Control users reached through other channels or devices dilute the measured lift.
• No other major changes in marketing or pricing during the test window.

When randomization is not possible, causal inference methods such as difference-in-differences or synthetic control can estimate lift from observational data, with more modeling assumptions.

Measuring incrementality with retailer or advertiser data in a clean room

Advertisers can measure incrementality with advertiser or retailer data, without exchanging customer-level records, by running the conversion or sales lift analysis inside a data clean room. Each party contributes its own data, the matching and calculation run in isolation, and only aggregated results leave.

The need arises because the two halves of a lift study sit with different companies. A publisher holds the exposure data, including which users were eligible for the ad and which were held back as a control. The advertiser or retailer holds the conversion or purchase data. The usual route is for the advertiser to send its conversion events or customer records to the platform, which is how lift tools inside walled gardens and other ad platforms work. Some advertisers cannot take that route. Regulated categories such as healthcare and financial services often restrict conversion data from going to an ad platform, and privacy regulation or internal governance policies can rule it out as well. In those cases, exposure data and conversion data have to be brought together without either party handing its records to the other.

A clean room study typically runs in this order:

1. The publisher or platform splits the target audience into a test group and a control group before the campaign starts.
2. During the campaign, the publisher provisions opportunity-to-see data for both groups into the clean room, typically updated daily. Only the test group's opportunities turn into delivered ads.
3. The advertiser or retailer brings its conversion or sales data into the same clean room, during or after the campaign.
4. The clean room matches the two datasets on hashed identifiers, such as hashed email addresses, and calculates lift from a pre-built template.
5. Each party receives the aggregated output: a match rate, the percentage lift between test and control, estimated incremental conversions, and a significance check. Results can be split by cohort, such as product line or region.
6. After the study, the clean room is shut down and both parties' data is deleted.

Privacy safeguards apply throughout. Results built from too small a group of users are withheld automatically before they leave the clean room. In a Decentriq data clean room, the analysis runs inside confidential computing, so neither the other party nor the operator sees the underlying records.

With a retailer, the same structure produces a sales lift read. The retailer's transaction or loyalty data supplies the outcome, including in-store purchases that online tracking never observes. In a Decentriq data clean room, Ad Alliance and OBI First Media Group matched OBI's CRM and retail media data with Ad Alliance's media and audience data for a GARDENA campaign, which reported an 8% uplift in ad recall and a 3% uplift in purchase intent.

Several conditions affect the quality of the result. Baseline tracking should be in place for a couple of weeks before the study starts, so test and control behavior can be compared on a like-for-like basis. The control group should not be heavily reached by other media outside the measured campaign, since that exposure blurs the comparison. The audience also needs to be large enough for results to clear significance and minimum-group thresholds.

Not every lift study needs a clean room. Platforms such as Amazon run experiments inside their own environments, because one company controls exposure and often has some view of conversion. A clean room becomes relevant when exposure data and conversion data sit with two separate companies that cannot hand raw records to each other. The conversion lift measurement guide walks through the study flow in more detail.

How to calculate incremental lift and iROAS

Four calculations turn the conversion rates and spend from a test into incrementality results:

• Relative lift: test group conversion rate ÷ control group conversion rate − 1
• Incremental conversions: (test group conversion rate − control group conversion rate) × number of users in the test group
• Share of conversions that were incremental: incremental conversions ÷ test group conversions
• Incremental ROAS (iROAS): incremental conversions × average order value ÷ ad spend
• Cost per incremental acquisition (CPIA): ad spend ÷ incremental conversions

The following illustration uses round numbers. A test group of 100,000 users converts at 2.0%, and a control group converts at 1.6%. Relative lift is 25%, and incremental conversions are (2.0% − 1.6%) × 100,000 = 400, which is 20% of the test group's 2,000 purchases. With an average order value of $50 and ad spend of $10,000, incremental revenue is $20,000 and iROAS is 2.0. The cost per incremental acquisition is $10,000 ÷ 400 = $25. A report that credits all 2,000 purchases to the campaign would show a return of 10.

The control group's 1.6% rate is the baseline: about 1,600 of the 2,000 purchases would have happened without the campaign. A significance check compares the two conversion rates and indicates whether the difference could be chance. Small groups or short tests widen the margin of error, and a real effect can then fail to clear the significance threshold.

Incrementality testing tools and when a clean room fits

The best tool for incrementality testing of media campaigns depends on the channel, on where the exposure and outcome data sit, and on the team running the test. Four categories cover most situations.

These categories complement each other. Test results from any of them can be used to calibrate attribution and MMM outputs, as described earlier.

When a clean room fits

A clean room suits a lift study in several situations:

• Exposure data sits with a publisher or platform, conversion data sits with the advertiser or retailer, and neither can hand over raw records.
• The advertiser operates in a regulated category, such as healthcare or financial services, where conversion data cannot be sent to an ad platform.
• The outcome is sales recorded by a retailer, including in-store purchases, and the lift must be matched at the customer level.
• The advertiser wants an independent read on performance, computed jointly with the media partner.

Other situations call for a different tool. Channels without user-level identifiers, such as TV and out-of-home, are better measured with geo tests. A study that stays inside one platform can use that platform's own lift tool.

When choosing a clean room provider for lift measurement, the level of service the provider offers is a main differentiator. Cloud-platform clean rooms such as Snowflake and AWS Clean Rooms supply the infrastructure, and study design, setup, and analysis typically fall to the advertiser's own data team. Providers also differ in whether an analysis requires custom SQL queries or runs from pre-built templates, which determines whether a marketing team can operate a study directly. Ownership of the operator is a second consideration. InfoSum is owned by WPP, and LiveRamp has agreed to be acquired by Publicis Groupe (announced in May 2026), so an advertiser measuring media bought through those groups should weigh that relationship. The Decentriq vs LiveRamp comparison covers these differences in more detail.

Decentriq runs conversion lift studies in its data clean room from pre-built templates, so a marketing team can operate a study without writing custom SQL. The media partner provisions test and control exposure data, the advertiser adds its conversion data, and the study returns a match rate, the lift, estimated incremental conversions, and a significance check, with optional breakdowns by cohort such as product line or region. Both parties' data is deleted when the study ends. Decentriq is not owned by an agency group or a media company, which keeps the measurement independent of the media being measured.

Frequently asked questions

What is incrementality in marketing and how do you measure it?

Incrementality is the additional outcome a marketing activity causes beyond what would have happened without it. It is measured with controlled experiments that compare an exposed test group with a held-out control group. The difference in outcomes, expressed as lift or incremental conversions, is the incremental effect.

How long does an incrementality test need to run?

Duration depends on conversion volume, the expected lift, and the test design. Tests commonly run for several weeks, and geo tests tend to need longer than user-level tests. Low conversion volumes require longer runs or larger audiences to reach statistical significance.

What are the best tools for incrementality testing of media campaigns?

The answer depends on the situation. Ad platform lift studies suit single-platform media, open-source libraries and dedicated platforms suit ongoing geo experiments, and data clean rooms suit studies in which exposure and conversion data sit with different companies. For the last case, Decentriq runs conversion lift studies from pre-built templates in its data clean room. Many teams use more than one category.

How can advertisers measure incrementality data when they cannot share customer data?

In a Decentriq data clean room, a publisher's exposure data and an advertiser's or retailer's conversion data are matched and compared inside a protected environment. Neither party sees the other's raw records, and only aggregated lift results are released, with small groups withheld. The conversion lift demo shows how a study runs.

Is a clean room required to measure incrementality?

No. Holdout, geo, and platform lift tests run without one. A clean room becomes useful when exposure data and conversion data are held by separate companies that cannot hand raw records to each other.